Table 2 Summary of network-only strategy
Task | Reference | Input | Output | Network | Training dataset | Loss function |
|---|---|---|---|---|---|---|
Dataset-driven (DD) approach | Sinha et al.114 | Diffraction image | Phase | U-Net and ResNet | Expt.: 10,000 pairs | l1-norm |
Li et al.115 | Diffraction image | Phase | U-Net and ResNet | Expt.: 10,000 pairs | NPCC | |
Deng et al.117 | Diffraction image | Phase | U-Net and ResNet | Expt.: 10,000 pairs | NPCC | |
Goy et al.118 | Weak-light diffraction | Phase | U-Net and ResNet | Expt.: 9500 pairs | NPCC | |
Wang et al.119 | In-line hologram | Phase | U-Net and ResNet | Expt.: 9000 and 11,623 pairs | l2-norm | |
Nguyen et al.120 | Multiple LR intensity images (Fourier ptychography) | HR phase | U-Net and DenseNet | Expt.: --- | GAN loss and l1-norm | |
Cheng et al.121 | LR intensity image (Fourier ptychography) | HR phase and amplitude | CNN and ResNet | Expt.: 20 fields-of-view | l2-norm | |
Cherukara et al.122 | Far-field diffraction | Phase or amplitude | SegNet (two) | Sim.: 180,000 pairs | Cross-entropy | |
Ren et al.123 | Off-axis hologram | Phase or amplitude | ResNet and SubPixelNet | Expt.: >10,000 pairs | l2-norm | |
Yin et al.124 | Hologram | Phase | U-Net | Expt.: 2400 and 200–2000 (unpaired) | Cycle-GAN loss | |
Lee et al.125 | Hologram | Phase and amplitude | U-Net and CNN | Expt.: 600–9060 (unpaired) | Cycle-GAN loss and SSIM | |
Hu et al.126 | Spots’ intensity image | Phase | U-Net and ResNet | Sim.: 46,080 pairs | l2-norm | |
Wang et al.127 | Defocus intensity image | Phase | U-Net and ResNet | Expt.: 20,037 pairs | l2-norm | |
Zhou et al.128 | LR defocus intensity image | HR phase | U-Net | Expt.: 1300 pairs | l2-norm | |
Pirone et al.129 | Hologram in different angles | Phase | CAN | Expt.: 4000 pairs | l1-norm | |
Chang et al.130 | Diffraction image (Electron) | Phase | U-Net and ResNet | Sim.: 250,000 pairs | l1-norm | |
Xue et al.132 | Bright- and dark-field images | Phase | U-Net and BNN | Expt.: 185 groups | l1-norm and uncertainty term | |
Li et al.133 | Two images of symmetric illumination | Phase | U-Net | Sim.: 1301 groups | GAN loss | |
Hologram | Phase and amplitude | Y-Net | Expt.: 1331 pairs | l2-norm | ||
Zeng et al.135 | Hologram | Phase or amplitude | CapsNet | Expt.: --- | l2-norm | |
Wu et al.136 | Far-field diffraction | Phase and amplitude | Y-Net | Sim.: 142,500 groups | Loss in real and reciprocal space | |
Huang et al.137 | Two or 3 holograms | Complex field | U-Net and Recurrent CNN | Expt.: 208 groups | GAN loss and l1-norm and SSIM | |
Uelwer et al.138 | Far-field diffraction | Phase | Cascaded neural network | Sim.: --- | l2-norm or l1-norm | |
Castaneda et al.139 | Off-axis hologram | Wrapped phase | U-Net | Expt.: 1512 pairs | GAN loss and TSM and STD | |
Jaferzadeh et al.140 | Off-axis hologram | Phase | U-Net | Expt.: 900 pairs | GAN loss | |
Luo et al.141 | Hologram | Phase | MCN | Expt.: 1 pair | Bucket error rate (BER) loss | |
Ding et al.142 | LR image | HR phase | U-Net and Swin Transformer | Expt.: 3500 and 3500 (unpaired) | Cycle-GAN loss | |
Ye et al.144 | Far-field diffraction | Complex field | MLP and CNN | Sim. and Expt.: --- | l1-norm | |
Three or 4 holograms | Complex field | ResNet and Fourier module (FIN) | Expt.: 600 groups | l1-norm, complex domain and perceptual loss | ||
Shu et al.147 | Hologram | Phase | Network based on NAS | Expt.: 276 pairs | MixGE and binary and sparsity loss | |
Physics-driven (PD) approach | Boominathan et al.149 | LR intensity images (Fourier ptychography) | HR Phase and amplitude | U-Net | Sim.: 1 (input only) | l2-norm with physical model |
Wang et al.150 | Diffraction image | Phase | U-Net | Sim. and Expt.: 1 (input only) | l2-norm with physical model | |
Zhang et al.151 | Diffraction image | Phase | U-Net | Sim. and Expt.: 1 (input only) | l2-norm with defocus distance and physical model | |
Diffraction image | Phase and amplitude | U-Net | Sim. and Expt.: 1–180 (input only) | l2-norm with aperture constraint and physical model | ||
Bai et al.154 | Hologram | dual-wavelength Phase | CDD | Expt.: 1 (input only) | l2-norm with physical model | |
Galande et al.155 | Hologram | Phase and amplitude | U-Net | Expt.: 1 (input only) | l2-norm with physical model and denoiser | |
Yao et al.159 | 3D diffraction image | Phase and amplitude | 3D Y-Net | Sim.: 52,000 (input only) | l2-norm with physical model | |
Li et al.160 | Two diffraction images | Phase | Two-to-one Y-Net | Sim.: 500 (input only) | l2-norm with physical model | |
Bouchama et al.161 | LR intensity images (Fourier ptychography) | HR Phase and amplitude | U-Net | Sim.: 10,000 (input only) | l2-norm with physical model | |
Huang et al.162 | Two holograms | Phase and amplitude | GedankenNet | Sim.: 100,000 (input only) | l2-norm and Fourier-domain l1-norm |
